Rank that relies on position alone, with no other visual cue, is easy to miss

What it is

Two modules look identical: same card, same title size, same button. The author says the left one is primary because “primaries live on the left.” In a test, nearly half cannot say which is primary, or pick the right-hand one. If rank is written only in coordinates, with no second perceptible difference, the position prior is often not enough. Position is not useless. Alone, the signal is weak: scan habits, the task, even which hand holds the phone can drown it.

Why it happens

The position prior is a statistical bias, not a high-contrast edge. Two objects tied on the feature maps give the selector no peak, so choice is near chance, or follows a habit orthogonal to the task (whichever is nearer the thumb; whichever just moved — and here nothing moved, so more chance). “Primary on the left” is the author’s production rule. It was never written into a visual channel, so observers cannot decode it. Stack even a cheap second cue — solid primary, ghost secondary, or one extra title size — and a peak appears; position can stay as weak redundant coding. Failure that “relied on position” looks like users “not looking the habitual way.” The code was never sent.

Studying it

Three sets: position only; position plus a size gap; position plus a luminance gap. Measure accuracy of “which is the primary” against a 50% baseline. If the position-only set sits near baseline and the stacked sets rise, position alone is insufficient. The effect is small, so n must be large; counterbalance left/right in a within-subjects design. Report landscape phones and left-handers separately; they change which side is near.

Where it stops holding

Strong platform convention can make position-only decodable: many people know, without reading style, that the start side of an iOS nav is Back. That is convention, not position physics, and it dies on another platform. When two objects sit far apart with nothing between them, the one met first will win and position-only looks sufficient — insert a third identical module in between and it is not. Row one versus row two in a list also differs in position, but list items usually also differ in index or content; a purely positional list barely exists.

Applying it

• When primary and secondary appear as a pair, give the primary a cue of its own besides the slot. Position is the duplicate, not the only channel.
• A comment that says “the left is primary” is not a distinction on the glass. Turn the comment into a visible difference.
• Under responsive layout, a left/right relation becomes top/bottom and a position-only primary/secondary is lost. One with a second cue survives.
• Check: a version where primary and secondary share a skin and only left/right remains. Ask five people which is primary. If hits are no better than chance, solid-fill the primary button or step the primary title one rung, and ask five others. A rise in accuracy means rank was sent. Do not explain a miss as “they didn’t read left to right” — that dumps an uncoded distinction on reading direction.

Related

• Same group: F3.09.1 The top and the start side are granted higher rank by default · F3.09.2 Positional rank can be overwritten by stronger visual weight · F3.09.3 The same position does not mean the same rank under every layout mode
• Nearby: F3.01.2 A difference on one dimension is easily cancelled by another · F3.02.1 A size difference must clear the discrimination threshold before it makes a level
• Search terms: positional hierarchy · redundant coding · visual weight · primary versus secondary